bug#60708: pcre: improve support for linking with a library without unicode

[Paul Eggert]

On 1/11/23 14:12, Carlo Arenas wrote:
> pcre2_config does a static check (defined at compile time) and
> therefore is unlikely to fail and might be even under the right
> circumstances optimized out.

Not sure what is meant by "static check" here. The call won't be 
optimized out unless you compile with -flto or equivalent, and have the 
source code to pcre2 as well as the source code to grep. And in that 
case the two forms should generate equivalent code (no insns needed).

> you are correct that setting the original value was meant to protect
> from that function failing and will ensure the original path was still
> being taken (which I thought was safer), while your suggested change
> will take the opposite one (not setting UTF in a multibyte locale,
> which will fail in different ways).

Oh, I think see your point, but doesn't this mean that even my code was 
too trusting? It should be something like this:

  if (localeinfo.multibyte)
    {
      uint32_t unicode;
      if (! (localeinfo.using_utf8
             && 0 <= pcre2_config (PCRE2_CONFIG_UNICODE, &unicode)
             && unicode))
        die (EXIT_TROUBLE, 0, _("-P supports only unibyte and UTF-8 
locales"));
      ...

That is, we're better off diagnosing the problem and not attempting to 
use pcre2 if the result will be wrong (or even result in undefined 
behavior). The problem is unlikely to occur so it's good to be 
conservative here.